Skip to content

Video

Seedance 1

Video

Available on Ofrai — chat, images, and video in one place with shared credits.

94 creditsModel details

Video studio

Preview

No clip yet

Your latest clip preview will show here after generation.

Open studio

Example prompts

Choose an example to load its prompt and recommended settings into the Seedance 1 creator above.

How it works

Create with Seedance 1 in three steps

  1. 01

    Choose your inputs

    Keep Seedance 1 selected, then add a prompt and any supported reference assets.

  2. 02

    Set the output

    Choose the aspect ratio, resolution, and available Video options before generating.

  3. 03

    Generate and refine

    Review the result, adjust the prompt or settings, and generate another version from the same place.

Doubao Seedance 1.0 Pro Quality is the reference-driven member of ByteDance's Doubao video line, and the documentation describes it as one endpoint that switches between two model ids — `doubao-seedance-1-0-pro-fast` and `doubao-seedance-1-0-pro-quality` — with the same parameter set underneath. It does text-to-video, and it accepts image control in three shapes: a first frame, a last frame, and a reference image.

That third shape is what separates it from most of the entries in this group, and it carries a documented price: when a reference image is in play, 1080p is removed from the resolution list. Duration runs from 2 to 12 seconds, the lowest floor of the four models here, and the aspect-ratio list is the widest, including the rarely used 21:9.

So the honest summary is: this is the model to reach for when a shot has to stay tied to supplied imagery, and the resolution ceiling it leaves behind is part of that deal rather than an accident.

One endpoint, two model ids, one shared parameter set

The interface description does something unusual: it defines a single Doubao SeeDance video-generation endpoint and says the model is selected through the `model` parameter, which takes either `doubao-seedance-1-0-pro-fast` or `doubao-seedance-1-0-pro-quality`. That framing matters more than it first appears, because it means the two ids are not documented as having different inputs or different envelopes — they are documented as two settings of the same request.

Every parameter discussed below is shared: text-to-video is supported by both, the image-control inputs are the same, duration is the same 2-to-12 range, and the aspect ratios are the same six. Generation is asynchronous, returning a task ID that you query for the result.

The practical read is that quality and speed are chosen at submission time rather than by rewriting your request, so the useful discipline is to draft and iterate against the fast id and switch to quality once the shot, the framing and the prompt are settled, rather than burning quality renders on prompts you have not finished writing.

The documented quirk: a reference image removes 1080p

This is the single most useful piece of information in the reference, and it is stated as a limitation rather than buried. The resolution parameter normally accepts three values — 480p, 720p and 1080p, with 720p as the default. But the documentation carries an explicit restriction: 1080p is not supported when using a reference image, specifically the `role: reference` slot inside `image_with_roles`. That is a real interaction, not a caveat about future support.

It means a project cannot have both identity fidelity and a full-HD render from this endpoint in a single request, and the choice has to be made deliberately: render the full-HD master from text or from first and last frames, and keep the reference-image version at 720p; or accept a reference-locked 720p take as the deliverable. Because the restriction is attached to one specific role rather than to image input in general, a first-frame or last-frame run keeps the full ladder.

Plan the two passes as different requests with different purposes, and the constraint stops being a limitation and becomes the reason you have a sharp version and a flexible version of the same shot.

Duration: a two-second floor is the distinguishing detail

The duration field is optional, documented values are 2, 5 and 12 seconds, and the stated supported range is 2 to 12 with 5 as the default. Of the four models described across this site, this is the only one that will render a two-second clip. That is not a trivial detail — two seconds is the length of a logo sting, a whip-pan insert, a match cut, a counter tick, or a beat between two shots in a fast montage.

Most text-to-video endpoints refuse those jobs outright, and the workaround is to generate something longer and trim it, which wastes render time on frames you will delete and gives the model room to drift in the part you keep. When a job genuinely is two seconds, this endpoint lets you say so. The same range also covers the ordinary case, so nothing about a longer shot is unusual here; the range is simply wider at the bottom.

The caution is the ordinary one: a single continuous take that has to hold a subject, a background and a camera move gets less reliable as it lengthens, so use the long end for shots that are genuinely continuous and cut anything else into shorter pieces.

Six aspect ratios, including the one nobody asks for

The aspect-ratio parameter defaults to 16:9 and accepts six values: 16:9 for landscape, 9:16 for vertical, 1:1 for square, 4:3 and 3:4 for the traditional ratios, and 21:9 for ultrawide. Five of those are unremarkable. 21:9 is not, and it is the one that makes this ladder worth checking against your brief rather than assuming.

An ultrawide request is the right call for a piece of landscape or architecture where the subject is small in a large environment, for a title sequence that needs to sit inside a scope frame, or for a horizontal composition that will be letterboxed rather than cropped — because the geometry is generated at the ratio you ask for, so asking for it here is not the same as cropping a 16:9 render down afterward.

The two traditional ratios matter for delivery pipelines built around older frame sizes, and asking for them at generation time avoids the pillarbox or crop that post-scaling introduces. Set the ratio when you submit, not after, and it stays consistent with a first-frame anchor rather than fighting it.

URL-only image input, and why the two list parameters never travel together

Image input is optional, and when supplied it is an array of image URLs used for image-to-video. The documentation is explicit that only URL format is supported and that base64 is no longer accepted — a specific, testable constraint that will save a debugging cycle. The URLs must be publicly reachable over http or https, and the reference points at an upload endpoint for turning local files into such a URL.

There is also a hard rule about which list you use: `image_urls` and `image_with_roles` cannot be sent in the same request. That is the second failure worth internalising, because the two look interchangeable and the error, if it surfaces at all, is about the request shape rather than the images. Decide which one carries your intent before you build the payload. `image_urls` is the plain first-frame array; `image_with_roles` is what gives you the role slots, including the reference role whose use costs you 1080p.

Mixing them in an attempt to pass a first frame and a reference together is the request this endpoint refuses.

Writing prompts that survive an image being attached

The prompt field is required on every request and is documented as a description of the video's content, with the guidance to describe scene, action and style in detail for better results. That guidance is written for the text-only case, and it needs one adjustment when an image is attached.

If a first frame is supplied, the model is already committed to what the opening looks like, and prose that re-describes visible content competes with the image rather than helping — a sentence about the subject's appearance is redundant, while a sentence about what happens next is not.

Reference images work the same way with one extra consequence: they are there to hold an identity, so the highest-value text is about behaviour and camera, not about who the subject looks like, because appearance is exactly what the reference is already supplying. A workable structure is scene, then action, then camera and pacing, then style words last so they read as a modifier.

The documented example is a beach sunset with golden light on the water and waves lapping the shore — specific, concrete, and about what moves — which is a fair template for the level of detail that helps here.

FAQ

Can I render 1080p while using a reference image?
No. The reference documents an explicit restriction: 1080p is not supported when a reference image is used — specifically the `role: reference` slot inside `image_with_roles`. First-frame and last-frame runs keep the full 480p/720p/1080p ladder. If you need both, plan two requests: a full-HD pass without the reference and a 720p reference-locked pass.
What durations can I request?
The supported range is 2 to 12 seconds with 5 as the default, and the documented values are 2, 5 and 12. A two-second floor is the distinguishing feature of this entry — it covers stings, whip-pan inserts and match cuts without generating a longer clip and trimming it.
Does the endpoint accept base64 images?
No. Only URL format is supported and base64 is no longer accepted. Use publicly reachable http or https URLs, or push a local file through the documented image-upload endpoint to obtain one.
Can I send image_urls and image_with_roles together?
No — the documentation states they cannot both be used in one request. `image_urls` is the plain first-frame array; `image_with_roles` is what exposes the role slots including the reference role. Pick the one that matches your intent before building the payload.

FAQ

Can I use Seedance 1 directly on this page?
Yes. The creator above is the working Ofrai studio with Seedance 1 selected by default.
Where are my finished results saved?
Signed-in generations are kept in your Ofrai library so you can download or revisit them.
Which prompts work best with Seedance 1?
Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
Can I edit the example prompts before generating?
Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.

Compare other Video models that can handle similar creative work to Seedance 1.

Commercial use depends on the selected model and model terms.